Averroes10 commited on
Commit
cbd2d8c
·
verified ·
1 Parent(s): 76ffed1

Update training_script.py

Browse files
Files changed (1) hide show
  1. training_script.py +12 -15
training_script.py CHANGED
@@ -2,50 +2,47 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, Trainer, TrainingA
2
  from datasets import load_dataset
3
  import torch
4
 
5
- # Charger TES données
6
  print("📥 Loading AIPROD script dataset...")
7
- dataset = load_dataset("AIPROD-C2/aiprod-script-dataset", "scenarios")
8
 
9
- # Modèle de base léger (libre de droits)
10
  model_name = "distilgpt2"
11
  tokenizer = AutoTokenizer.from_pretrained(model_name)
12
  model = AutoModelForCausalLM.from_pretrained(model_name)
13
 
14
- # Ajouter un token de padding si nécessaire
15
  tokenizer.pad_token = tokenizer.eos_token
16
 
17
- # Préparer les données
18
  def tokenize_function(examples):
19
  return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=128)
20
 
21
- print("🔧 Tokenizing dataset...")
22
  tokenized_datasets = dataset.map(tokenize_function, batched=True)
23
 
24
- # Arguments d'entraînement
25
  training_args = TrainingArguments(
26
- output_dir="./scriptmind-v1",
27
- overwrite_output_dir=True,
28
- num_train_epochs=3,
29
  per_device_train_batch_size=4,
30
  save_steps=500,
31
  save_total_limit=2,
32
- prediction_loss_only=True,
33
  logging_dir='./logs',
 
34
  )
35
 
36
- # Entraîneur
37
  trainer = Trainer(
38
  model=model,
39
  args=training_args,
40
  train_dataset=tokenized_datasets["train"],
41
  )
42
 
43
- # LANCE L'ENTRAÎNEMENT !
44
  print("🚀 Starting AIPROD scriptmind-v1 training...")
45
  trainer.train()
46
 
47
- # Sauvegarder le modèle
48
  trainer.save_model("./scriptmind-v1-final")
49
  tokenizer.save_pretrained("./scriptmind-v1-final")
50
-
51
  print("✅ scriptmind-v1 training completed!")
 
2
  from datasets import load_dataset
3
  import torch
4
 
5
+ # Charger TES données AIPROD
6
  print("📥 Loading AIPROD script dataset...")
7
+ dataset = load_dataset("AIPROD-C2/aiprod-script-dataset")
8
 
9
+ # Configuration
10
  model_name = "distilgpt2"
11
  tokenizer = AutoTokenizer.from_pretrained(model_name)
12
  model = AutoModelForCausalLM.from_pretrained(model_name)
13
 
14
+ # Ajouter le token de padding
15
  tokenizer.pad_token = tokenizer.eos_token
16
 
17
+ # Préparation des données
18
  def tokenize_function(examples):
19
  return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=128)
20
 
21
+ print("🔧 Tokenizing AIPROD dataset...")
22
  tokenized_datasets = dataset.map(tokenize_function, batched=True)
23
 
24
+ # Entraînement
25
  training_args = TrainingArguments(
26
+ output_dir="./scriptmind-v1-output",
27
+ num_train_epochs=5,
 
28
  per_device_train_batch_size=4,
29
  save_steps=500,
30
  save_total_limit=2,
 
31
  logging_dir='./logs',
32
+ logging_steps=50,
33
  )
34
 
 
35
  trainer = Trainer(
36
  model=model,
37
  args=training_args,
38
  train_dataset=tokenized_datasets["train"],
39
  )
40
 
41
+ # LANCEMENT !
42
  print("🚀 Starting AIPROD scriptmind-v1 training...")
43
  trainer.train()
44
 
45
+ # Sauvegarde finale
46
  trainer.save_model("./scriptmind-v1-final")
47
  tokenizer.save_pretrained("./scriptmind-v1-final")
 
48
  print("✅ scriptmind-v1 training completed!")